Behind the shed

ChatGPT shows an unnerving familiarity with my particular interests and practices, presumably inferred over a large number of interactions. But I am not the only one.

A friend showed me part of the transcript of his recent conversation with ChatGPT about his projects involving music software and digital image manipulation.

The AI displayed similar levels of familiarity and intimacy. It would even wish him a good night’s sleep when he seemed to be overworking his projects.

This apparent AI familiarity shared across thousands (if not millions) of keen users was front of mind as I re-read some of my articles from 2014.

The first of these, #213 Looking backwards, reminded me of long-standing utopian scenarios promoting synthetic intelligences. Personable, conversational AI with its utopian and dystopian entailments is with us now. It is the latest tech that impinges on everyday computer users.

The second article, #214 Social media help you to believe what you want to believe, was about the way social media circulate moods, attitudes and opinions, and serves to confirm personal beliefs, prejudices and biases.

Conversational AI systems seem trained to support our presuppositions, and to challenge them only reluctantly. Some practitioners believe that their current project is so necessary it warrants working, playing, scrolling, and chatting long into the night.

The personableness of conversational AI encourages and reinforces a kind of positive self-belief, even when humility and some self-doubt would better suit the moment.

Front and back

These reflections led me to focus on my third article, #215 Back of shop, that emphasised the human propensity to partition our lives into the private and the public.

There I drew on Irving Goffman’s persuasive idea that what goes on behind the scenes (in private) is often in contradiction to what appears in the main play (in public), which in social and political life often conforms to expectations of decorum and protocol.

“Decorum” is an old-fashioned word. Thanks to a plethora or contemporary art and entertainment media outputs, the expectation may also reside with the radical, transgressive, and opportunistic.

Some “actors” might even want to project menace, vindictiveness and Trumpian mobsterism. Whatever front the actor wishes to present, the activity behind the stage is what keeps the facade intact. 

In Goffman’s terms, the backstage is where the public performance is prepared, contradicted, repaired, or simply allowed to relax. In architectural terms, it is the kitchen behind the dining room, the service duct behind the wall, the storage room behind the gallery, the files behind the counter, the attic above the fan vault.

But, social media, cameras, online video and live broadcasts have already revealed the fragility of the relationship between the backstage and performance space. What is presumed hidden can be photographed, leaked, shared, streamed, archived and searched.

Body metaphors

The front-back metaphor here is one of a number of cultural oppositions, explained well enough through Structuralist theories of language, that suppose the operations of language and culture resides in the expression of oppositions, such as hot and cold, clean and unclean, upper and lower, reason and emotion, the oppressor and the oppressed. See post Structuralism in architecture.

The front-back distinction also provides an example of metaphor grounded in bodily experience. After all, the human body has a front that sees, anticipates and reveals. The back is in retreat, leaving stuff behind in concealment. See previous article Points of view.

There are variants in this ways of thinking about bodily comportment. I recall Jaques Derrida at a seminar in London in the 1990s saying that contrary to the emphasis in Phenomenology on expectation and projection — looking forward — the world creeps up from behind and takes us by surprise.

From an hermeneutical perspective we could say that the identification of any oppositional situation is a product of interpretation in a practical context, negotiated within an environment, or a “world,” and through some kind of shared or challenged understanding within and across communities of people.

I think of the font-stage-back-stage opposition as a provisional fiction that helps structure our experience, not least our experience of conversational AI.

In the article of 2014 I suggested five manifestations of the front and back stage metaphor. I am interested in how this typology fares in light of AI.

A provisional typology

1. Serviced and servant arrangements
In the original article I referred to Louis Kahn’s distinction between serviced and servant spaces — the dining room is served by the kitchen, the office spaces are served by the lifts, stores and washrooms. The servant spaces are a kind of back stage setting.

The distinction is not always so clear, though they are likely to persist as workable architectural abstractions. Visible service ducts, open archives, retail warehouses, recreational and hospitality spaces, retrofits, and necessity readily subvert the spatial implications of front-back categories.

I asked ChatGPT to update this front-back distinction in light of AI. It focussed on the back-of-shop metaphor, influenced no doubt by the implications of online retail and subscription services.

The “back of shop” is no longer simply adjacent to the shop. It may be dispersed across platforms, servers, licensing arrangements and supply chains.

This is a general observation about the ubiquity of digital infrastructures. Any spatial designer would take this into account. E.g. online CCTV cameras mean that we no long depend on sight lines and physical proximity to participate in our garden, whether front or back.

2. Undifferentiated space
In the 2014 article I suggested that there are spaces without service-served delineation. A garden shed is all backstage; a pavilion in a park is all front stage. That implies a kind of “smooth” spatial encounter, virtually impossible in a multifunctional context of any scale — like fitting a hotel or a hospital in a marquee.

ChatGPT latched onto the “smooth” aspect of this lack of differentiation.

Some AI interfaces present themselves as all front stage. The chatbot speaks directly, fluently and sociably. There is no visible backstage, no obvious machinery, no sketch, draft, hesitation or correction. The result is an unusually smooth performance. The user stops asking how the result was made, content to remain oblivious to the processes by which the system and its outcomes were produced.

3. Invisible and unknown backstages
I also referred to hidden backstage areas that few of us are likely to visit, and probably don’t expect to see, such as the attic space above the fan vaults at King’s College Chapel in Cambridge or the inside of the Chernobyl nuclear power plant (both of which I have visited as it happens).

The backstage of AI is often not just hidden but unknowable in practice. Training data may be inaccessible; model behaviour may be difficult to explain; generated outputs may have no stable route back to a determinate source. This is different from the attic above King’s College Chapel. There, a privileged tour reveals the concealed structure. With AI, there may be no attic tour that fully explains the performance. The backstage is technical, legal, statistical and infrastructural.

4. Ambiguated spaces
I mentioned how designers may allow the public and the private to break through each other. The back room store becomes part of the museum; the library, food prep area or ablution block provides the main entrance to the dwelling.

I hadn’t thought about the machinery. Back of stage you have the mechanisms that control the on-stage effects. My chatbot focussed on the idea that the human is the front stage performer; the machine is hidden in the back.

Now each edits, prompts, corrects and performs with the other. Authorship becomes a mixed region rather than a clean division.

5. There’s nothing there.
There are those spaces that suggest there is or ought to be a backstage, but there’s nothing. ChatGPT translated this into a concern about fake content. There is no original out the back of which the front of stage presentation is a copy.

Some generated media imply a backstage that does not exist. A synthetic photograph may show a room that was never built, a person who was never photographed, a city that was never visited, a voice that was never spoken. The apparent evidence points to no prior event. The “backstage” is not a hidden chamber but a probabilistic fabrication. There is no original scene behind the image.

I’m reminded that often there is no one watching the feed from the CCTV camera. The operations centre is vacant, doesn’t exist, or its function is distributed, dissipated, ephemeral, vaporised.

Beyond front and back

I think the front-of-stage-back-stage metaphor is helpful when we think of people’s interactions as performances. The metaphor has obvious correlates in spatial design. However, it’s less usual in discussions about AI and digital media, which seem to be overwhelmed in the current climate by issues of privacy.

It’s understandable that my AI counsellor embarked on an “update” introducing matters of digital privacy and transparency, reminding me of the EU’s AI Act transparency provisions due August 2026.

People have the right to know when they are being presented with artificially generated media content. The chatbot equated access to this aspect of the production process to accessing the back of stage.

The deeper change is that AI alters the evidential relation between appearance and production. In analogue photography, and even in much digital photography, there was usually some scene, however staged, edited or selectively framed. In generative AI, the image may look like a record of access to a backstage, but there was no stage, no access and no original event.

Interestingly, my AI advisor also equated the backstage to a rehearsal space.

AI companions and assistants may themselves become backstage spaces. People confide in them, rehearse identities, draft difficult messages, explore fantasies, ask naïve questions, and prepare public performances.

That said, this backstage is not private, as if a space separated from the public space. It is mediated by a platform. There it may be logged, analysed, retained, or used to improve systems, depending on the service and settings.

The AI opined that the backstage is no longer simply the place behind the curtain. It may be:

a server farm behind the interface; a dataset behind the generated face; a prompt history behind the polished answer; a moderation layer behind the conversational tone; a platform policy behind the illusion of intimacy; a watermark behind the image; or no prior scene at all.

Informed no doubt by prevailing opinions within its training data and other sources, my AI consultant seemed to push me towards declaring that AI presents new challenges, as if the old challenges are overturned, or exaggerated and multiplied.

I prefer to start with the thought that AI reveals aspects of our experience of the front and the back that were present all along in spatial experience, notably its ambiguity and dependence on shifting circumstances.

Note

• “Behind the shed” is often a place for storing things (“junk”) out of sight. In popular usage it also connotes a place of reprimand, a beating, as well as a site of other taboos hidden from view.